Image Retrieval System using Semi-Automatic Relevance Feedback through 3D Model

Abstract

This paper studies and develops image retrieval system using semi-automatic method. The Self-organizing map (SOM) is applied to classify data and obtain 3D data model for user interface. This model reduces user workload in relevance feedback process because it classifies data into separate groups those allow users to input only on the representative model for effective learning. This system implements relevance feedback model using a network of radial basis function (RBF) which provides nonlinear discriminant analysis for measurement of image similarity. The RBF performs better than linear and quadratic functions. In a retrievalcycle, the suitable parameters input to RBF function are obtained from fed-back samples through 3D model. From the experiment results, the semi-automatic retrieval system implementing 3D model offered accurate retrieval results and reduced user workload as compared to the traditional relevance feedback methods.